Progress on the QUDA code suite

Ron Babich; Richard Brower; Mike Clark; Steven Gottlieb; Balint Joó; Guochun Shi

Progress on the QUDA code suite

Ron Babich, Richard Brower, Mike Clark, Steven Gottlieb, Balint Joó, Guochun Shi

Research output: Contribution to journal › Conference article › peer-review

Abstract

At the time of Lattice 2010, we were about to announce a distribution of the code (QUDA 0.3) that supported both Wilson/clover and improved staggered quarks for computation on a single GPU. Multi-GPU code was running for both solvers, but with the restriction of grid partitioning in only the time dimension. In the past year, we developed code that allows us to cut the lattice in all four dimensions. This allows us to scale computations to order 100 GPUs yielding multi-teraflop performance. We will present results for both types of solvers on GPU clusters and for other kernels important for physics projects. We also compare performance and cost-effectiveness of full application codes running on CPUs with our GPU accelerated code.

Original language	English
Journal	Proceedings of Science
Volume	139
State	Published - 2011
Externally published	Yes
Event	29th International Symposium on Lattice Field Theory, Lattice 2011 - Squaw Valley, Lake Tahoe, United States Duration: Jul 10 2011 → Jul 16 2011

Funding

We are grateful to LLNL (Edge), NERSC (Dirac), FNAL (J/Psi) and TACC (Longhorn) where we ran benchmarks. This work was supported in part by NSF grants OCI-0946441, OCI-1060012, OCI-1060067, and PHY-0555234, as well as DOE grants DE-FC02-06ER41439, DE-FC02-06ER41440, DE-FC02-06ER41443, DE-FG02-91ER40661, and DE-FG02-91ER40676. BJ additionally acknowledges support under DOE grant DE-AC05-06OR23177, under which Jefferson Science Associates LLC manages and operates Jefferson Lab. GS is funded through the Institute for Advanced Computing Applications and Technologies (IACAT) at the University of Illinois at Urbana-Champaign.

Funders	Funder number
Institute for Advanced Computing Applications and Technologies
National Science Foundation	OCI-0946441, PHY-0555234, OCI-1060012, OCI-1060067
U.S. Department of Energy	DE-FG02-91ER40676, DE-FC02-06ER41443, DE-FG02-91ER40661, DE-FC02-06ER41440, DE-AC05-06OR23177, DE-FC02-06ER41439
University of Illinois at Urbana-Champaign

Cite this

@article{d2c64ce45ee74c16adbcec56fc5ba2c2,

title = "Progress on the QUDA code suite",

abstract = "At the time of Lattice 2010, we were about to announce a distribution of the code (QUDA 0.3) that supported both Wilson/clover and improved staggered quarks for computation on a single GPU. Multi-GPU code was running for both solvers, but with the restriction of grid partitioning in only the time dimension. In the past year, we developed code that allows us to cut the lattice in all four dimensions. This allows us to scale computations to order 100 GPUs yielding multi-teraflop performance. We will present results for both types of solvers on GPU clusters and for other kernels important for physics projects. We also compare performance and cost-effectiveness of full application codes running on CPUs with our GPU accelerated code.",

author = "Ron Babich and Richard Brower and Mike Clark and Steven Gottlieb and Balint Jo{\'o} and Guochun Shi",

note = "Publisher Copyright: {\textcopyright} Copyright owned by the author(s).; 29th International Symposium on Lattice Field Theory, Lattice 2011 ; Conference date: 10-07-2011 Through 16-07-2011",

year = "2011",

language = "English",

volume = "139",

journal = "Proceedings of Science",

issn = "1824-8039",

publisher = "Sissa Medialab",

}

TY - JOUR

T1 - Progress on the QUDA code suite

AU - Babich, Ron

AU - Brower, Richard

AU - Clark, Mike

AU - Gottlieb, Steven

AU - Joó, Balint

AU - Shi, Guochun

PY - 2011

Y1 - 2011

N2 - At the time of Lattice 2010, we were about to announce a distribution of the code (QUDA 0.3) that supported both Wilson/clover and improved staggered quarks for computation on a single GPU. Multi-GPU code was running for both solvers, but with the restriction of grid partitioning in only the time dimension. In the past year, we developed code that allows us to cut the lattice in all four dimensions. This allows us to scale computations to order 100 GPUs yielding multi-teraflop performance. We will present results for both types of solvers on GPU clusters and for other kernels important for physics projects. We also compare performance and cost-effectiveness of full application codes running on CPUs with our GPU accelerated code.

AB - At the time of Lattice 2010, we were about to announce a distribution of the code (QUDA 0.3) that supported both Wilson/clover and improved staggered quarks for computation on a single GPU. Multi-GPU code was running for both solvers, but with the restriction of grid partitioning in only the time dimension. In the past year, we developed code that allows us to cut the lattice in all four dimensions. This allows us to scale computations to order 100 GPUs yielding multi-teraflop performance. We will present results for both types of solvers on GPU clusters and for other kernels important for physics projects. We also compare performance and cost-effectiveness of full application codes running on CPUs with our GPU accelerated code.

UR - http://www.scopus.com/inward/record.url?scp=85053332463&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85053332463

SN - 1824-8039

VL - 139

JO - Proceedings of Science

JF - Proceedings of Science

T2 - 29th International Symposium on Lattice Field Theory, Lattice 2011

Y2 - 10 July 2011 through 16 July 2011

ER -

Progress on the QUDA code suite

Abstract

Funding

Other files and links

Fingerprint

Cite this